EECS Seminar: On-device Contextual AI – Challenges and Opportunities

Abstract: In this talk Vikas Chandra will present the challenges and opportunities in the area of on-device contextual AI. While Large language models and Large Vision-language models are making significant progress on a weekly basis, the next frontier is on-device where these models can run locally in real time with a lot more context than just vision and language. This talk will discuss broad aspects of contextual AI, outlining research opportunities, as well as delve deeper in Meta’s recent work on enabling small language models (MobileLLM) and efficient vision-language models (MiniGPT-v2).

Bio: Vikas Chandra is senior director at Meta Reality Labs where he works on AI research focusing on AR and VR products. Prior to Meta, he was director of Applied Machine Learning at Arm Inc. He received his Ph.D. and M.S. degrees in electrical and computer engineering from Carnegie Mellon University. He held the positions of visiting scholar (2011-2014) and visiting faculty (2016-2017) in the EE department at Stanford University. He has authored 120+ research publications and is an inventor on 40+ US and international patents. Chandra received the ACM-SIGDA Technical Leadership Award in 2009 and was invited to the 2017 Frontiers of Engineering Symposium organized by the National Academy of Engineering. He is a senior member of IEEE.